Hamza H.

Internal AI marketing-intelligence SaaS platform (TypeScript, Next.js 16, React 19).

Multi-tenant AI platform: architected Karmable AI on a 47-table PostgreSQL schema (Drizzle ORM) where autonomous agents run on cron schedules inside isolated cloud sandboxes, with organization- and project-scoped data isolation enforced on every query.

Provider-agnostic LLM layer: built a unified LLM gateway (Vercel AI SDK) routing across Claude, GPT, Gemini, and Perplexity with custom token-metering middleware for per-call billing, and bridged ~30 external APIs as tools via Model Context Protocol (MCP) and Composio (500+ integrations).

Multi-agent shared memory: designed a 3-tier knowledge protocol (atomic facts → observations → reflections) backed by vector and hybrid (BM25 + semantic) search, so agents reuse each other's research across runs.

38 AI tools, 16 autonomous agents: implemented agents for ICP generation, competitor discovery, SEO/SERP analysis, Reddit and social listening, and content drafting, plus a streaming chat UI with artifacts (code/text/sheet/image) and an AI reporting engine with 22+ chart types.

Billing & observability: engineered an immutable, idempotent credit-billing ledger fed by Stripe subscriptions and on-chain USDC micro-payments (x402, Base L2, viem); instrumented full observability (Sentry, Langfuse, OpenTelemetry) with Playwright and Vitest tests, Docker, and GitHub Actions CI/CD.

Integriverse & NetSuite: delivered Integriverse, an AI workflow-automation product, and an AI-driven pipeline that automates schema mapping, transformation, and synchronization into NetSuite with human-in-the-loop review gates.

AtomAI (neuro-symbolic): built a hybrid engine pairing deterministic ANTLR parsing with RAG‑augmented LLMs to generate syntactically valid VBA macros, eliminating hallucinated code in manufacturing automation (FastAPI, SQL Server, ChromaDB).

Code validation & tooling: engineered a Semantic Analyzer that validates AI‑generated code against a reflected API model before execution, and shipped PDMShell, a CLI bringing legacy SOLIDWORKS PDM systems into modern CI/CD pipelines.

• • AI workflow orchestrator: Architected a GCP orchestrator for property-management operations; built a Map-Reduce engine on Cloud Run (Terraform IaC) for parallel batch processing of LLM reasoning tasks and a hybrid gRPC + batch API core sustaining 99.9% uptime.

• • Human-in-the-loop & security: Built a React/FastAPI dashboard for operators to pause, resume, and audit AI decisions in real time, with a zero-trust layer (per-task AES-256 encryption, key expiry, PII redaction).